Comparative Study of Residential Energy Disaggregation Techniques from Low Frequency Sampling Measurements

Abstract

In this article, techniques for residential energy disaggregation are compared to identify and separate the consumption of different appliances. Traditional methods such as Combinatorial Optimization (CO) and Factorial Hidden Markov Models (FHMM) are contrasted with Deep Learning (DL) approaches like Recurrent Neural Networks (RNN) and Denoising Autoencoders (DAE), using data from the UK-DALE and DATAPORT databases. The results highlight the superiority of DL techniques in accuracy, despite the additional training time and computational resources required. Transfer Learning (TL) also poses challenges in generalizing models to different geographical contexts. This article underscores the importance of considering not only accuracy but also computational cost when choosing energy disaggregation methods, contributing to research in residential energy efficiency and emphasizing the need for new strategies.